Signal

What it is

Signal (the repo is named vetcall-insights) is an automated platform that listens to every phone call that comes into a DE client practice and tells you what happened. Every call from CallRail — the phone-tracking platform that all DE practices already use — flows into Signal, where an AI reads the transcript and asks a precise set of questions: Did the caller book an appointment? If not, why not? What was the urgency? What is the estimated revenue that just left? That analysis gets packaged into a daily morning email for each practice — one version for the front desk, one for the practice manager — and into a web dashboard where anyone can dig deeper. Signal currently covers 28 active practices generating roughly 167 calls per day, with 180 practices loaded in the system and ready to activate.

Why it exists

DE's clients pay for marketing but rarely engage with the raw data their marketing generates. CallRail produces transcripts and recordings that almost no one reads — the data is too dense, the volume too high, and the useful signal buried. That leads to churn: practices do not see the value, so they leave. Signal solves this by collapsing a day's worth of calls into one clear, actionable morning report. The goal is not just retention — it is to show practices the dollar value of calls they are losing so they call those people back. The math is stark: a practice missing 15 to 20 appointments a month at $200 a visit is losing $3,000 to $4,000 monthly. Even recovering 25 percent of that pays for the product 2 to 7 times over.

How it works

Signal runs on a daily three-step pipeline, kicked off automatically each morning by a scheduled job. Step one is ingestion: at 8 AM UTC, the system pulls every call from the previous day out of CallRail's API — auto-paginating through all 250-call pages — and stores each call in the database, including the full transcript and recording link. CallRail also pushes calls in real time via webhooks roughly 20 minutes after each call ends, so the database is actually being updated throughout the day, not just overnight; the morning batch is a safety net that catches anything the webhooks missed.

Step two is analysis. Four batches of calls are processed by the AI across the morning, with each batch sending transcripts to Claude Haiku — Anthropic's fast, low-cost model — through a technique called tool-calling. Instead of asking the AI to write a free-form summary, Signal gives it a strict form to fill out: 27 named fields, each with defined options. Things like disposition (Answered, Voicemail, Abandoned), appointment booked (Yes or No), missed opportunity type (None, Price Shopping, Fully Booked, Agent Did Not Offer, Referred, or Other), a one-line summary, a coaching tip for the staff member, and a ready-to-use callback script. The AI's output costs roughly half a cent per call.

Step three is compilation. At 11 AM UTC, the day's analyses are rolled up into a report for each practice: total calls, booking rate, missed revenue, and a callback queue sorted from highest dollar value to lowest. That report drives two emails — one for the front desk with a focused list of who to call back plus scripts, and one for the practice manager with the full performance picture. Both land before 7 AM practice time. A web dashboard lets anyone log in and explore history, trends, and individual call detail including transcripts and recordings.

The parts

Where it fits